带聚集的移动方差

Question

我有大约6分钟的频率热电偶数据。热电偶安装在不同的高度，每一高度都有一个按径向位置区分的热电偶。

DT_TI_RECORDED      HEIGHT      POS             TEMPERATURE
2018-05-16 00:00:00     1       90              111
2018-05-16 00:00:00     1       180             112
2018-05-16 00:00:00     1       270             113
2018-05-16 00:00:00     2       90              112
2018-05-16 00:00:00     2       180             114
2018-05-16 00:00:00     2       270             115
2018-05-16 00:00:00     3       90              112
2018-05-16 00:00:00     3       180             112
2018-05-16 00:00:00     3       270             113
...
2018-05-16 00:06:00     1       90              111
2018-05-16 00:06:00     1       180             112
2018-05-16 00:06:00     1       270             113
2018-05-16 00:06:00     2       90              112
2018-05-16 00:06:00     2       180             114
2018-05-16 00:06:00     2       270             112
2018-05-16 00:06:00     3       90              114
2018-05-16 00:06:00     3       180             112
2018-05-16 00:06:00     3       270             114
...

每6分钟，对于每一个独特的高度和位置组合，我想计算一个向后的n小时移动方差，让我们说4小时。

我试图复制的原始代码是为SAS stats包编写的。

    PROC EXPAND DATA=Raw_data
        OUT=Moving_Variance
        ALIGN = BEGINNING
    ;
    by HEIGHT POS;
    ID DT_TI_RECORDED ;
        CONVERT TEMPERATURE = Moving_4hour_Var /  METHOD = none TRANSFORMOUT = (MOVVAR 40); 
    #/* 40 obs at 6min freq = 4hour moving variance*/
    QUIT;

我花了几个小时搜索谷歌，我想我需要使用的R库叫做zoo，我想要的函数是rollapply，但是我不知道如何将聚合和rollapply组合起来。

我试过了

moving_var <- Raw_data %>%
              aggregate(HEIGHT,POS) %>%
              rollapply( TEMPERATURE, width = 40, FUN = sd, fill = NA)

但不起作用。我对R编程非常陌生，这让我发疯了。

Answer 1

尝试以下聚合：

library(zoo)

result = aggregate(temp ~ pos + height,
              data = df,
              FUN = function(x){
                  rollapply(x, width = 40, FUN = var, by = 40)
              }
)

width是滚动窗口的宽度，而by是下一个窗口起点跳过的点数。每个窗口中有40个，您将在前一个窗口的末尾得到每个窗口的开始。

生成的数据框架对于每个窗口都有一个列。这一结构可视为“宽”结构。如果您希望将其作为“长”格式，请使用来自tidyr的gather或来自reshape2的melt。

示例：

df = structure(list(pos = c(0, 90, 180, 270, 0, 90, 180, 270, 0, 90, 
                            180, 270, 0, 90, 180, 270, 0, 90, 180, 270, 0, 90, 180, 270, 
                            0, 90, 180, 270, 0, 90, 180, 270, 0, 90, 180, 270, 0, 90, 180, 
                            270, 0, 90, 180, 270, 0, 90, 180, 270, 0, 90, 180, 270, 0, 90, 
                            180, 270, 0, 90, 180, 270, 0, 90, 180, 270, 0, 90, 180, 270, 
                            0, 90, 180, 270, 0, 90, 180, 270, 0, 90, 180, 270, 0, 90, 180, 
                            270, 0, 90, 180, 270, 0, 90, 180, 270, 0, 90, 180, 270, 0, 90, 
                            180, 270, 0, 90, 180, 270, 0, 90, 180, 270, 0, 90, 180, 270, 
                            0, 90, 180, 270, 0, 90, 180, 270, 0, 90, 180, 270, 0, 90, 180, 
                            270, 0, 90, 180, 270, 0, 90, 180, 270, 0, 90, 180, 270, 0, 90, 
                            180, 270, 0, 90, 180, 270, 0, 90, 180, 270, 0, 90, 180, 270, 
                            0, 90, 180, 270), height = c(1L, 1L, 1L, 1L, 2L, 2L, 2L, 2L, 
                            3L, 3L, 3L, 3L, 4L, 4L, 4L, 4L, 1L, 1L, 1L, 1L, 2L, 2L, 2L, 2L, 
                            3L, 3L, 3L, 3L, 4L, 4L, 4L, 4L, 1L, 1L, 1L, 1L, 2L, 2L, 2L, 2L, 
                            3L, 3L, 3L, 3L, 4L, 4L, 4L, 4L, 1L, 1L, 1L, 1L, 2L, 2L, 2L, 2L, 
                            3L, 3L, 3L, 3L, 4L, 4L, 4L, 4L, 1L, 1L, 1L, 1L, 2L, 2L, 2L, 2L, 
                            3L, 3L, 3L, 3L, 4L, 4L, 4L, 4L, 1L, 1L, 1L, 1L, 2L, 2L, 2L, 2L, 
                            3L, 3L, 3L, 3L, 4L, 4L, 4L, 4L, 1L, 1L, 1L, 1L, 2L, 2L, 2L, 2L, 
                            3L, 3L, 3L, 3L, 4L, 4L, 4L, 4L, 1L, 1L, 1L, 1L, 2L, 2L, 2L, 2L, 
                            3L, 3L, 3L, 3L, 4L, 4L, 4L, 4L, 1L, 1L, 1L, 1L, 2L, 2L, 2L, 2L, 
                            3L, 3L, 3L, 3L, 4L, 4L, 4L, 4L, 1L, 1L, 1L, 1L, 2L, 2L, 2L, 2L, 
                            3L, 3L, 3L, 3L, 4L, 4L, 4L, 4L), temp = 1:160), .Names = c("pos", 
                            "height", "temp"), row.names = c(NA, -160L), class = "data.frame")

> head(df,20)
   pos height temp
1    0      1    1
2   90      1    2
3  180      1    3
4  270      1    4
5    0      2    5
6   90      2    6
7  180      2    7
8  270      2    8
9    0      3    9
10  90      3   10
11 180      3   11
12 270      3   12
13   0      4   13
14  90      4   14
15 180      4   15
16 270      4   16
17   0      1   17
18  90      1   18
19 180      1   19
20 270      1   20


library(zoo)

result = aggregate(temp ~ pos + height,
              data = df,
              FUN = function(x){
                  rollapply(x, width = 3, FUN = var, by = 3)
              }
)

将导致：

   pos height temp.1 temp.2 temp.3
1    0      1    256    256    256
2   90      1    256    256    256
3  180      1    256    256    256
4  270      1    256    256    256
5    0      2    256    256    256
6   90      2    256    256    256
7  180      2    256    256    256
8  270      2    256    256    256
9    0      3    256    256    256
10  90      3    256    256    256
11 180      3    256    256    256
12 270      3    256    256    256
13   0      4    256    256    256
14  90      4    256    256    256
15 180      4    256    256    256
16 270      4    256    256    256